Tuberkuloz ve toraks · 2025
Segmentation of honeycomb cysts, traction bronchiectasis and emphysematous lung parenchyma using the deep learning method.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Aydin Nevin, Yildirim Hüseyin, Alataş Füsun, Sariay Büşra, Mert Burcu, Demir Serhat, Çelik Özer, Aslan Ahmet Faruk, Odabaş Alper
Аннотация
INTRODUCTION: This study aimed to perform the segmentation of honeycomb cysts, traction bronchiectasis, and emphysematous lung parenchyma in highresolution computed tomography (HRCT) examinations using the deep learning method of artificial intelligence.
MATERIALS AND METHODS: The study included the cross-sectional images of 265 patients diagnosed with usual interstitial pneumonia between 2017 and 2021. Minimum intensity projection (MinIP) was performed on axial sections in the parenchymal window. Emphysema areas, traction bronchiectasis, and honeycomb cysts were segmented and labeled on the axial HRCT images. The dataset was divided into three parts, namely training, validation, and testing, at a ratio of 80, 10, and 10%, respectively. The results were calculated by selecting 50% as the threshold value for the intersection over union (Jaccard index) statistic.
RESULT: Of the 265 patients included in the study, 184 (69.4%) were male and 81 (30.6%) were female. Mean age of the patients was 73 ± 10 years. In the test group segmented as emphysema, the sensitivity, precision, F1 score, area under the curve (AUC), and accuracy values were calculated as 0.81, 0.90, 0.85, 0.86, and 0.75, respectively. In the test group segmented as traction bronchiectasis, the sensitivity, precision, F1 score, AUC, and accuracy values were found to be 0.95, 0.76, 0.85, 0.80, and 0.75, respectively. In the test group segmented as honeycomb cysts, the sensitivity, precision, F1 score, AUC, and accuracy values were 0.96, 0.92, 0.94, 0.88, and 0.90, respectively.
CONCLUSIONS: In this study, we successfully utilized the U-Net architecture, a deep learning technology, to accurately segment honeycomb cysts in MinIP images, one of the parameters that will help classify interstitial lung diseases (ILDs). We anticipate that our study will guide future studies on the classification of ILDs.
MATERIALS AND METHODS: The study included the cross-sectional images of 265 patients diagnosed with usual interstitial pneumonia between 2017 and 2021. Minimum intensity projection (MinIP) was performed on axial sections in the parenchymal window. Emphysema areas, traction bronchiectasis, and honeycomb cysts were segmented and labeled on the axial HRCT images. The dataset was divided into three parts, namely training, validation, and testing, at a ratio of 80, 10, and 10%, respectively. The results were calculated by selecting 50% as the threshold value for the intersection over union (Jaccard index) statistic.
RESULT: Of the 265 patients included in the study, 184 (69.4%) were male and 81 (30.6%) were female. Mean age of the patients was 73 ± 10 years. In the test group segmented as emphysema, the sensitivity, precision, F1 score, area under the curve (AUC), and accuracy values were calculated as 0.81, 0.90, 0.85, 0.86, and 0.75, respectively. In the test group segmented as traction bronchiectasis, the sensitivity, precision, F1 score, AUC, and accuracy values were found to be 0.95, 0.76, 0.85, 0.80, and 0.75, respectively. In the test group segmented as honeycomb cysts, the sensitivity, precision, F1 score, AUC, and accuracy values were 0.96, 0.92, 0.94, 0.88, and 0.90, respectively.
CONCLUSIONS: In this study, we successfully utilized the U-Net architecture, a deep learning technology, to accurately segment honeycomb cysts in MinIP images, one of the parameters that will help classify interstitial lung diseases (ILDs). We anticipate that our study will guide future studies on the classification of ILDs.